Connect and run your agent in Slack

Run any supported agent framework in Slack through a cloud-hosted Intelligence connection.

Run your agent as a Slack app. CopilotKit Intelligence holds the Slack credentials, and your process holds the agent and the application logic.

New to the product? Start with the Channels overview to understand how the SDK, Runtime, Intelligence, and provider connection fit together.

Before you continue, configure the Channel in Intelligence. You should have CHANNEL_CODE and CPK_INTELLIGENCE_API_KEY.

Start with your coding agent

Use this prompt to connect your agent to Slack, start the runner, and verify a real workspace message. You can also follow the manual steps below.

Open your coding agent in your project's folder, or in an empty folder for a new app.

Before you start

  • Node.js 22 or later; the managed launcher requires the global WebSocket available in Node.js 22+
  • A long-running Node host or container; serverless request handlers cannot own the persistent gateway connection

Build and run your Channel

Create the runner

Terminal
mkdir my-slack-channel
cd my-slack-channel
npm init -y
npm pkg set type=module

Use this tested package combination. Keep exact versions in your lockfile and verify compatibility before changing either package.

Terminal
npm install --save-exact @copilotkit/channels@0.11.0 @copilotkit/runtime@1.73.3
Terminal
npm install -D tsx typescript @types/node

Add a NodeNext TypeScript configuration:

tsconfig.json
{
  "compilerOptions": {
    "target": "ES2022",
    "module": "NodeNext",
    "moduleResolution": "NodeNext",
    "strict": true,
    "skipLibCheck": true,
    "noEmit": true,
    "types": ["node"]
  },
  "include": ["*.ts", "*.tsx"]
}

Connect your agent backend

The runner can use any supported backend. The selector's Agent backend controls the setup below. It must export a fresh makeAgent(threadId) result for each Slack conversation; do not share one stateful agent instance across threads.

Declare the managed Slack Channel

Create the runner below. Replace support-slack with the exact Code shown in Intelligence.

Set agent: makeAgent on the Channel because this handler calls thread.runAgent(). Channels do not inherit runtime.agents. A Channel whose handlers reply directly without runAgent() can omit agent.

channel.ts
import { createServer } from "node:http";
import { createChannel } from "@copilotkit/channels";
import {
  CopilotKitIntelligence,
  CopilotRuntime,
} from "@copilotkit/runtime/v2";
import { createCopilotNodeListener } from "@copilotkit/runtime/v2/node";
import { makeAgent } from "./agent.js";

function required(name: string): string {
  const value = process.env[name];
  if (!value) throw new Error(`Missing required environment variable: ${name}`);
  return value;
}

const channel = createChannel({
  name: required("CHANNEL_CODE"),
  identifyUser: "platform",
  agent: makeAgent,
});

channel.onMessage(async ({ thread, message }) => {
  await thread.runAgent({
    context: [
      { description: "Originating platform", value: message.platform },
    ],
  });
});

const intelligence = new CopilotKitIntelligence({
  apiKey: required("CPK_INTELLIGENCE_API_KEY"),
  // Cloud-hosted deployments use the default URLs. Override both URLs
  // together only for self-hosted or non-production Intelligence.
  apiUrl: process.env.INTELLIGENCE_API_URL,
  wsUrl: process.env.INTELLIGENCE_GATEWAY_WS_URL,
});

const runtime = new CopilotRuntime({
  agents: {},
  intelligence,
  channels: [channel],
});

// Wire teardown before the listener exists, because creating it is what
// starts the Channel. A Ctrl-C during the connect window then still tears
// the Channel down instead of hitting Node's default handler.
let teardown: (() => Promise<void>) | undefined;
const shutdown = async () => {
  await teardown?.();
};
process.once("SIGINT", shutdown);
process.once("SIGTERM", shutdown);

const listener = createCopilotNodeListener({
  runtime,
  basePath: "/api/copilotkit",
});
const channels = listener.channels;
const server = createServer(listener);
teardown = async () => {
  await channels.stop();
  if (server.listening) server.close();
};

// Optional: block startup until the activation above settles, so a broken
// deploy fails loudly instead of serving as a bot that never answers.
await channels.ready({ timeoutMs: 30_000 });
const status = channels.status();
if (status.overall !== "online") {
  throw new Error(`Slack Channel is not online: ${JSON.stringify(status)}`);
}

const port = Number(process.env.PORT ?? 3001);
server.listen(port, () => {
  console.log(`Slack Channel online; lifecycle server listening on :${port}`);
});

Calling createCopilotNodeListener starts the Channel. ready() is optional. It waits until the start settles and throws if it fails. ready() can settle while setup is still incomplete, so check status() before you report the Channel online. Without ready(), start failures appear in your logs.

Configure secrets and start

.env
CPK_INTELLIGENCE_API_KEY=<project-api-key>
CHANNEL_CODE=support-slack
# Add the agent variables shown for your selected backend.
PORT=3001

# Optional paired overrides for self-hosted or non-production Intelligence:
# INTELLIGENCE_API_URL=https://intelligence.example.com
# INTELLIGENCE_GATEWAY_WS_URL=wss://realtime.intelligence.example.com

The runner uses port 3001 so it can run alongside a web app on port 3000. If 3001 is already in use, set PORT to another free port.

Where these values come from

Three variables matter here:

  • OPENAI_API_KEY, or the key for whichever model provider your agent uses. This one is yours to create and paste.
  • CPK_INTELLIGENCE_API_KEY: npx copilotkit@latest project select writes it to .env and provisions the project and its key.
  • CHANNEL_CODE: the Code shown on the Channel in Intelligence.

If you started with the coding-agent prompt above, the onboarding run writes both Intelligence values for you. To set them yourself, see configure the runtime handoff for where each value appears in Intelligence.

Cloud-hosted Intelligence supplies both default base URLs. For a self-hosted or non-production deployment, override both together. The REST and realtime planes use separate hosts, so do not derive the WebSocket URL from the API URL. Pass each as a bare base URL without /api, /socket, /runner, or /client. Create the project-scoped runtime key from API Keys in the Intelligence project sidebar.

  1. Add .env to .gitignore.

  2. Start the selected agent backend.

  3. Start the runner:

    Terminal
    node --env-file=.env --import tsx channel.ts

Intelligence should change from Waiting for runtime to Online.

Know the healthy state

StatusWhat to check
DisabledEnable the Channel before expecting delivery.
Setup incompleteFinish the selected provider's required setup fields before starting the runtime.
Setup failedReopen platform setup and correct the rejected credentials or configuration.
Waiting for runtimeStart the process, and match its Code and provider to this Channel.
ConflictCompare every replica's complete Channel declaration set. Identical replicas should elect one active owner and connected standbys; different but overlapping sets are unsafe.
OfflineCheck the runner process, network, and Intelligence gateway connection.
Delivery failingCheck platform credentials, app permissions, and the provider response.
OnlineThe runtime is connected; send a real provider message to verify the full path.

Verify a real Slack message

Invite the app if needed, then mention it in a channel:

Slack
/invite @your-app
@your-app summarize the decisions in this thread

Also test a direct message. A response in the real workspace validates the Slack token pair, Intelligence delivery, the runner, the AG-UI agent, and reply path end to end.

Tool-call progress

Managed Slack keeps tool-call progress out of the streamed reply by default, so the conversation ends with a clean result. Tool lifecycle events still remain in Intelligence history and are available during replay.

Opt in per Channel when the live tool timeline is useful:

channel.ts
const channel = createChannel({
  name: required("CHANNEL_CODE"),
  identifyUser: "platform",
  agent: makeAgent,
  showToolStatus: true,
});

Troubleshooting

Next, map application users and choose Memory grants, add tools and context, or build interactive approvals.

How is this guide?